UAV Pilot Identification using Machine Learning Techniques - Research Proposal

The topic of this research proposal is UAV pilot identification using machine learning techniques.The research hypothesis of the project is that by using biometric data related to individual pilot style, based on measurements from radio controlled (RC) signals and the on-board IMU sensor, it is possible to reduce the risk of UAVs being hijacked and used in illegal activities.

-Chapter 1 . Firstly, an introduction to UAVs and their applications is needed, demonstrating the importance of the application area. Next, the specific security issues need to be introduced, and why they pose a challenge. The chapter could then commence by identifying biometrics and pilot identification as an interesting direction to be pursued, bringing on board the use of machine learning techniques. The project aim and objectives will then need to be introduced. Here, the security issues, specific scenarios and pilot statuses need to be clarified, and then, the objectives should be clearly planned to lead to delivering the overall aim of the research. Finally, the chapter should provide an overview of what is to follow in Chapters 2, 3, etc.

-Chapter 2. This chapter should focus on background material. The first part should be focused on the workings of UAVs, providing the fundamental technological framework of a UAV (in terms of navigation, control and communications). Each of the sections describing aspects of UAVs should be backed up by equations and references to state-of-the-art published work. Following the introduction to UAVs, the second section of the Chapter should very clearly elaborate on security concerns and how they have been addressed. The first part of this section should focus on cybersecurity approaches, and the second part on machine learning ones. There should be a critical analysis of the techniques presented in terms of concept/philosophy, methods, scenarios to be used, advantages and disadvantages for each of the state-of-the-art published work, which is considered. The second part of the section should focus on machine learning approaches, following the same format as the previous section, i.e., concept/philosophy, methods, scenarios to be used, advantages and disadvantages for each of the state-of-the-art published work. An important aspect of the research is also allied research in biometrics for non-UAV applications, e.g., biometrics for human drivers, etc. A detailed state-of-the-art review should be performed here.

Chapter 3 should focus on the techniques to be used. Specifically, given the subject matter of machine learning, and the problem of class imbalance, which is of real concern in this application, the chapter should start by defining machine learning, classifying machine learning techniques into categories, before providing detailed scientific descriptions of the various algorithms to be used (including figures and basic equations). Performance metrics of classifier systems need to be introduced. Next, focus should shift on class imbalance, in terms of defining the problem.

A section on the impact of class imbalance in machine learning should be included, before an extensive scientific description of the various types of techniques used to address it. Given the requirements of the project, and the specific characteristics of the datasets, a critical analysis of the features of various classifiers should be carried out, selecting with justification a number of classifiers for further analysis

Chapter 4 should focus on the preliminary results. First, the two datasets should be clearly introduced and analysed. Next, the chapter should branch out to experiments on RC signals and provide the specific implementation information (e.g., number of hidden layers, and hidden units per layer in the case of a MLP), which would be required in order to replicate the research.

A variety of performance metrics should be used to analyse the performance of individual classifiers. The same approach should be followed the second dataset, i.e., RC and IMU data, while focusing on implementation details and a variety of performance metrics. A discussion section will bridge the results obtained for the two datasets and potentially raise opportunities for further development.

Chapter 5 should provide the concise conclusions of the research proposal. First, a summary of the outcomes of each of the chapters should be given. Then focus should shift on discussing and analysing the preliminary results obtained. The last section of this chapter is future work, which should be based on the outcomes of the results obtained, by referring to approaches in the published literature

 1- What is the relevance of the explanative machine learning intended for the random forest (RF)?  

2- What is the relevance of the cross-validation for the random forest algorithm? The the RF performs this implicitly.  

3- The protocol of experimentation need to be thoroughly revised and justified.

4- The drone data which is a set of continuous signals seem to be fed in its raw form to the classifier.

This might raise questions about the validity of this option in terms of efficiency and accuracy. There is need to elaborate and explain further this aspect.